MMOOZZGG.com

Calculator

Cost per token on your own server: what a million tokens costs

4 min read10 sources

The cloud sells tokens at a price per million, and you can calculate the cost per token on your own server in the same unit. The calculator takes your monthly costs and output speed, accounts for utilization and shows your price per million tokens next to cloud model prices as of 29 September 2026.

On this page · 4 sections

Calculator

What a million tokens cost on your own hardware

Your price

  • Tokens per month64.6 million
  • Dollars per million tokens$5.88

Single-stream measurements, in tokens per second: a 7B model runs at 83.1 on an M4 Max and 119.9 on an M5 Max, an 8B model at 127.7 on an RTX 4090 and 144.5 on an H100, and a 70B one at 19–25.

Cloud price per million output tokens, $

  • GPT-6 Luna$0.50
  • Russian provider B, Lite model, sync$0.77
  • Russian provider A, Lite model, async$1.18
  • Gemini 3.1 Flash-Lite$1.50
  • Russian provider A, Lite model, sync$2.37
  • Gemini 3.8 Flash (promo until 31 Dec 2026)$3.75
  • Russian provider B, Max model, async$3.85
  • GPT-5.4 mini$4.50
  • Claude Haiku 4.5$5.00
  • Russian provider B, Pro model, sync$5.92
  • Russian provider A, Pro model, async$7.23
  • Russian provider B, Max model, sync$7.70
  • Claude Sonnet 5.5$10.00
  • GPT-6 Sol$10.00
  • Gemini 3.1 Pro (up to 200k)$12.00
  • Russian provider A, Pro model, sync$14.22
  • Claude Opus 5.5$20.00
  • GPT-5.5$30.00
  • Claude Fable 5.1$50.00
  • GPT-6 Astra$50.00

Rows that are not dimmed cost more than your price per million.

Defaults come from the sources as of 29 September 2026. Dollar figures use the official US dollar rate set by the Bank of Russia, the country's central bank: 84.41. The estimate is approximate; every field can be edited.

The formula

Tokens per month: speed in tokens per second, multiplied by 2,592,000 seconds in a month and by the share of time the model actually generates. Price per million: the monthly costs from the local AI calculator, divided by the number of millions. Take the speed from measurements: for one user it is 19–145 tokens per second on models of 7–70 billion parameters. For batch output on a server card the total is, by our estimate, tens of times higher, and this is the main lever.

Examples at prices as of 29 September 2026

A Mac Studio M4 Max at $400 a month, a 7-billion model (Llama 2 7B in 4 bits), 83 tokens per second, generating 30% of the time: 64.6 million tokens and $6 per million, about $5.9. A workstation with an RTX 5090 at $800 a month, 128 tokens per second by the RTX 4090 measurement on an 8-billion model (the sources have no RTX 5090 measurement), 50% of the time: 165 million tokens, $5 per million. A 70-billion model in 4 bits takes about 38.5 GB and does not fit on one RTX 5090 with 32 GB. A workstation with two RTX 5090 cards costs about $1,100 a month, and at 19 tokens per second, by the measurement of two RTX 4090 cards, that is 24.8 million tokens and $46 per million, more than the output tokens of Opus 5.5 at $20. A rented H100 at $3 per hour on a 70-billion model at 25 tokens per second and 80% utilization: $41 per million; on an 8-billion model at 144.5 tokens per second, $7. A server with four H100 cards at $6,100 a month and batch output of 2,000 tokens per second in total (our assumption) at 60% utilization: 3.1 billion tokens and $2 per million, about $2.

What follows from this

One user on a local 7–8 billion model pays $5 to $6 per million tokens. That is the level of mid-range cloud models for output tokens, Haiku 4.5 at $5 and GPT-5.4 mini at $5, and 9–12 times more than cheap ones, GPT-6 Luna at about $0.50. Batch output for dozens of users on a server card brings the price down to $2 in the example with four H100 cards. That is cheaper than mid-range cloud models for output tokens, with full control over the data. A 70-billion model for one user costs $40 to $43 per million, like the cloud flagships: companies choose it for the data, not for the price.

How to measure your own speed

Rent the card you need for an hour: an RTX 4090 from about $0.50, an H100 from $3 per hour. Install an inference engine and run 50 of your own requests at 1, 4 and 16 parallel streams: the total speed at 16 streams is the number for the calculator. On a Mac, measure in LM Studio: it shows the speed after each answer.

Questions and answers

Why 30–60% utilization and not 100%?

The model waits for requests at night and between dialogs. By an estimate in an industry publication, the token price stays reasonable at a utilization of at least 80–90%, and without a task queue such utilization is hard to reach.

Should input tokens be counted?

Processing the request is much faster than generation: on an M5 Max it is 3,220 tokens per second against 120, 27 times faster. So in a rough estimate, input tokens are almost free. For long documents, measure the request processing time on your own card and add it to the utilization.

Sources and fact-checking

Checked by the editors on 29 September 2026.

  1. llama.cpp, performance on Apple Silicon, table updated 25 August 2026 (github.com)
  2. GPU-Benchmarks-on-LLM-Inference, Llama 3 8B and 70B, May 2024 (github.com)
  3. Intelion Cloud, H100/H200 rental (checked 29 September 2026) (intelion.cloud)
  4. Tproger, cloud GPUs for ML in 2026, 21 May 2026 (tproger.ru)
  5. Claude Platform, API prices (checked 29 September 2026) (platform.claude.com)
  6. Puter, review of OpenAI API prices based on openai.com, 29 September 2026 (developer.puter.com)
  7. ComNews, local AI and the cost of automation, 19 February 2026 (comnews.ru)
  8. GPU Index, RTX 4090 rental in Russia: 173 offers, from about $0.40 per hour (checked 30 September 2026) (gpuindex.ru)
  9. GPU Index, H100 rental in Russia: 41 offers, from $3 per hour (checked 30 September 2026) (gpuindex.ru)
  10. Regard, GeForce RTX 5090: one card on sale, 30 September 2026 (regard.ru)

Found a mistake? Tell us

Publications

Contact us